A Thorough Evaluation of Distance-Based Meta-Features for Automated Text Classification

Canuto, Sérgio D.; Sousa, Daniel Xavier; Gonçalves, Marcos André; Rosa, Thierson Couto

doi:10.1109/tkde.2018.2820051

Cited by 28 publications

(17 citation statements)

References 20 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Figure 1 shows the MetaLazy steps to predict a test instance t. For every test instance (1) MetaLazy selects some features (2) and creates new co-occurrence features (3). After that, MetaLazy selects the k most similar instances to t in the training set (4). In (5) the distances are used to create weights for each neighbor of t. And finally, in (6), MetaLazy fits a weaker classifier to predicts t.…”

Section: Weaker Classifier Selectionmentioning

confidence: 99%

“…For topic categorization, we have the following benchmark datasets: (1) 20 Newsgroups (20ng); (2) 4 Universities (4UNI), a.k.a. WebKB; (3) ACM Digital Library (ACM); (4) and Reuters90. For the sentiment analysis task, we have Yelp Reviews, which consists of a set of business and services reviews and Twitter Stanford, an automatically created dataset based on a set of known positive and negative emoticons.…”

Section: Datasetsmentioning

confidence: 99%

“…For the SVM (The multi-class SVM strategy applied was one-against all configured with nested stratified cross-validation within the training set) and Logistic Regression, we varied both the norm used in the penalization (l1,l2) and the penalty parameter C (0.10, 0.1, 10, 25). For BERT, we used the tunning suggested by the authors of the method: batch size (16,32), learning rate with Adam (5e-5, 3e-5, 2e-5), and the number of epochs (3,4,5). For the CNN, we varied optimizer learning rate (0.01, 0.001, 0.0001), activation function (relu, linear), optimizer (SGD, Adam and RMSprop), strides (1,2), kernel size (3,4,5), regularization (l1,l2 and l1l2) and pooling type (max and average pooling).…”

Section: Baselinesmentioning

confidence: 99%

“…Recent years have glimpsed a steady development in text representations (e.g., embeddings [20], metafeatures [4]) and text mining/classification algorithms [3], including deep learning solutions [10]. Such developments are mostly supported by a "tripod" constituted of: (1) vast amounts of (sometimes proprietary) data for training or vocabulary construction; (2) exploration of massive computational power, usually available only to large corporations; (3) increasingly complex learning models such as deep neural networks with many layers and intricate architectures.…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

"Keep it Simple, Lazy" -- MetaLazy: A New MetaStrategy for Lazy Text Classification

Mendes

Gonçalves

Cunha

et al. 2020

Proceedings of the 29th ACM International Conference on Information &Amp; Knowledge Management

View full text Add to dashboard Cite

Recent advances in text-related tasks on the Web, such as text (topic) classification and sentiment analysis, have been made possible by exploiting mostly the "rule of more": more data (massive amounts) more computing power, more complex solutions. We propose a shift in the paradigm to do "more with less" by focusing, at maximum extent, just on the task at hand (e.g., classify a single test instance). Accordingly, we propose MetaLazy, a new supervised lazy text classification meta-strategy that greatly extends the scope of lazy solutions. Lazy classifiers postpone the creation of a classification model until a given test instance for decision making is given. MetaLazy exploits new ideas and solutions, which have in common their lazy nature, producing altogether a solution for text classification, which is simpler, more efficient, and less data demanding than new alternatives. It extends and evolves the lazy creation of the model for the test instance by allowing: (i) to dynamically choose the best classifier for the task; (ii) the exploration of distances in the neighborhood of the test document when learning a classification model, thus diminishing the importance of irrelevant training instances; and (iii) a better representational space for training and test documents by augmenting them, in a lazy fashion, with new co-occurrence based features considering just those observed in the specific test instance. In a sizeable experimental evaluation, considering topics and sentiment analysis datasets and nine baselines, we show that our MetaLazy instantiations are among the top performers in most situations, even when compared to state-of-the-art deep learning classifiers such as Deep Network Transformer Architectures. CCS CONCEPTS • Computing methodologies → Machine learning algorithms; • Information systems → Clustering and classification.

show abstract

Section: Weaker Classifier Selectionmentioning

confidence: 99%

Section: Datasetsmentioning

confidence: 99%

Section: Baselinesmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

"Keep it Simple, Lazy" -- MetaLazy: A New MetaStrategy for Lazy Text Classification

Mendes

Gonçalves

Cunha

et al. 2020

Proceedings of the 29th ACM International Conference on Information &Amp; Knowledge Management

View full text Add to dashboard Cite

show abstract

“…1,2 Recently, many works have been proposed for the TC. Canuto et al 3 presented a use of a multiobjective optimization strategy to reduce the number of metafeatures while maximizing the classification effectiveness. Do and Poulet 4 proposed a novel fast and accurate parallel local support vector machine (SVM) algorithm for classifying very high-dimensional input spaces and large-scale multiclass data sets.…”

Section: Introductionmentioning

confidence: 99%

An improved term weighting scheme for text classification

Zhong

2019

Concurrency and Computation

View full text Add to dashboard Cite

Summary Text representation is a necessary and primary procedure in performing text classification (TC), which first needs to be obtained through an information‐rich term weighting scheme to achieve higher TC performance. So far, term frequency‐inverse document frequency (TF‐IDF) is the most widely used term weighting scheme, but it suffers from two deficiencies. First, the global weighting factors IDF in TF‐IDF approaches infinity if a certain term does not occur in a text. Second, the IDF is equal to zero if a certain term appears in any text. To offset these drawbacks, we first conduct an in‐depth analysis of the current term weighting schemes, and subsequently, an improved term weighting scheme called term frequency‐inverse exponential frequency (TF‐IEF) and its various variants are proposed. The proposed method replaces IDF with the new global weighting factor IEF to characterize the global weighting factor log‐like IDF in the corpus, which can greatly reduce the effect of feature (term) with high local weighting factor TF in term weighting. As a result, a more representative feature can be generated. We carried out a series of experiments on two commonly used data sets (corpora) utilizing Naïve Bayes and support vector machine classifiers to validate the performance of our proposed schemes. Experimental results explicitly reveal that the proposed term weighting schemes come with better performance than the compared schemes.

show abstract

Meta-learning of Text Classification Tasks

Madrid

2019

Progress in Pattern Recognition, Image Analysis, Computer Vision, and Applications

View full text Add to dashboard Cite

A Thorough Evaluation of Distance-Based Meta-Features for Automated Text Classification

Cited by 28 publications

References 20 publications

"Keep it Simple, Lazy" -- MetaLazy: A New MetaStrategy for Lazy Text Classification

"Keep it Simple, Lazy" -- MetaLazy: A New MetaStrategy for Lazy Text Classification

An improved term weighting scheme for text classification

Meta-learning of Text Classification Tasks

Contact Info

Product

Resources

About